A train circuit breaker intelligent control system

Abstract

This invention belongs to the field of intelligent control technology for train circuit breakers, and discloses an intelligent control system for train circuit breakers. The system includes: an acquisition module for acquiring electrical, environmental, and status parameters in response to circuit breaker tripping; a prediction module for obtaining the false trip probability through a non-faulty tripping identification model; and a processing module for processing the trip based on the false trip probability, which can trigger delayed reset, forced disconnection, or output prompts. This system utilizes a multi-layer model to accurately identify false tripping of circuit breakers when trains are operating in cold regions, improving the intelligence and reliability of train circuit breaker control and effectively ensuring the stable operation of the electrical system of trains in cold regions.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for train circuit breakers, specifically to an intelligent control system, method, electronic device, medium, and computer program product for train circuit breakers. Background Technology

[0002] Trains operating in high-altitude and cold regions, especially during the high-humidity and low-temperature periods of spring and autumn, may experience non-faulty tripping (false tripping) of high-voltage circuit breakers due to condensation entering the circuit breaker or surrounding control circuits. This type of problem is often misidentified as a load short circuit or equipment aging, leading to frequent power outages by the circuit breaker. Traditional train circuit breaker control systems rely heavily on single parameters (such as sudden current changes or arc signals) to diagnose these problems, lacking comprehensive analytical capabilities.

[0003] Therefore, there is an urgent need for an intelligent control system for train circuit breakers to improve the intelligence and reliability of train circuit breaker control and effectively ensure the stable operation of the electrical system of trains in harsh environments such as high-altitude and cold regions. Summary of the Invention

[0004] In response, the present invention provides an intelligent control system, method, electronic device, medium, and computer program product for train circuit breakers, to at least partially solve the above-mentioned technical problems.

[0005] This invention provides an intelligent control method for train circuit breakers, comprising the following steps:

[0006] In response to a circuit breaker trip, relevant parameters of the circuit breaker are acquired, including electrical parameters and environmental parameters. These parameters are then input into a non-fault trip identification model. Based on this model, the false trip probability of the circuit breaker is predicted. The non-fault trip identification model includes an input layer, a feature extraction layer, and a fusion decision layer. The input layer receives the relevant circuit breaker parameters, the feature extraction layer outputs environmental and electrical feature vectors, and the fusion decision layer outputs the false trip probability. Based on the false trip probability, the circuit breaker trip is processed, including triggering a delayed reset, forcibly disconnecting, or outputting a processing prompt.

[0007] In another aspect, this application also provides an intelligent control system for train circuit breakers, comprising:

[0008] A first acquisition module is used to acquire relevant parameters of the circuit breaker in response to circuit breaker tripping. The relevant parameters include electrical parameters, environmental parameters, and status parameters. A first prediction module is used to input the relevant parameters of the circuit breaker into a non-fault tripping identification model and predict the false tripping probability of the circuit breaker based on the identification model. The non-fault tripping identification model includes an input layer, a feature extraction layer, and a fusion decision layer. The input layer is used to receive the relevant parameters of the circuit breaker. The feature extraction layer outputs environmental feature vectors and electrical feature vectors. The fusion decision layer outputs the false tripping probability. A processing module is used to process the circuit breaker tripping based on the false tripping probability, including triggering a delayed reset of the circuit breaker, forcibly disconnecting, or outputting a processing prompt.

[0009] In another aspect, this application also provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the intelligent control method for train circuit breakers as described above.

[0010] In another aspect, this application provides a computer-readable storage medium having stored thereon computer program instructions that can be executed by a processor to implement the intelligent control method for train circuit breakers as described above.

[0011] Another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the intelligent control method for a train circuit breaker as described above.

[0012] The solution provided in this application utilizes a deep learning model to semantically discriminate circuit breaker tripping behavior, introduces environmental state simulation and condensation inference mechanisms, and reasonably predicts local environmental risks. It is particularly suitable for the proactive identification and avoidance of electrical malfunctions in high-altitude, cold and humid regions (such as the Qinghai-Tibet Plateau and tunnel entrances and exits). Through malfunction identification and very short-term prediction, dynamic handling of circuit breaker tripping behavior is achieved, including strategies such as tripping delay, secondary confirmation, and manual intervention, providing a flexible response mechanism. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0014] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0015] Figure 1 This is a schematic diagram of an intelligent control method for a train circuit breaker provided in an embodiment of the present invention.

[0016] Figure 2 This is a schematic diagram of the non-fault trip identification model provided in an embodiment of the present invention.

[0017] Figure 3 This is a schematic diagram of the post-trip processing flow of a circuit breaker provided in an embodiment of the present invention.

[0018] Figure 4 This is a schematic diagram of the structure of an intelligent control system for a train circuit breaker provided in an embodiment of the present invention.

[0019] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In a typical configuration of this application, the terminal and the service network devices each include one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0022] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, for example, read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0023] Computer-readable media include permanent and non-permanent, removable and non-removable media, which can store information by any method or technology. Information can be computer program instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, read-only optical disc (CD-ROM), digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0024] like Figure 1 As shown in the diagram, an embodiment of the present invention discloses a schematic diagram of an intelligent control method for a train circuit breaker, including the following method steps:

[0025] Step S100: In response to the circuit breaker tripping, obtain relevant parameters of the circuit breaker, including electrical parameters and environmental parameters.

[0026] In some embodiments, the electrical parameters of the circuit breaker include, but are not limited to, current, voltage, and frequency. Specifically, current transformers, voltage sensors, and other devices are deployed in the main circuit and related lines of the circuit breaker to collect current and voltage data. For example, the current may fluctuate between 100-300A when the train is running, and the sensor collects 1,000 data points per second. The frequency of the power system is monitored by a frequency sensor. In addition, the circuit breaker itself is equipped with a monitoring module to record information such as the time of change of opening and closing status and the duration of the action. For example, the timestamp of the moment of closing is accurate to milliseconds.

[0027] The environmental parameters of a circuit breaker mainly refer to the local environmental parameters and external environmental parameters of the circuit breaker's location. For example, temperature and humidity composite sensors and dew point sensors are installed on the surface of the circuit breaker enclosure and inside the equipment compartment to monitor ambient temperature, relative humidity, and dew point temperature in real time. Simultaneously, an infrared thermometer detects the surface temperature of the enclosure. The dew point temperature is the critical temperature at which water vapor in the air begins to condense into liquid water (dew). When the ambient temperature is lower than the dew point temperature, water vapor in the air will condense on the surface of objects, leading to a decline in the performance of insulation materials (such as moisture absorption of internal metal components and insulation parts of the circuit breaker), and consequently causing false tripping. Furthermore, in scenarios such as trains passing through tunnels or rainy weather, humidity will rise rapidly, affecting the dew point temperature; therefore, obtaining the temperature and humidity outside the train is also necessary.

[0028] Step S200: Input the circuit breaker-related parameters into a pre-trained non-fault tripping identification model, and predict the false tripping probability of the circuit breaker based on the identification model. The non-fault tripping identification model includes an input layer, a feature extraction layer, and a fusion decision layer. The input layer is used to receive the circuit breaker-related parameters. The feature extraction layer outputs environmental feature vectors and electrical feature vectors. The fusion decision layer outputs the false tripping probability.

[0029] Optionally, the feature extraction layer includes an environmental feature extraction module and an electrical feature extraction module, wherein the environmental feature extraction module adopts a CNN-LSTM structure and outputs an environmental feature vector; the electrical feature extraction module adopts an LSTM structure and outputs an electrical feature vector; the fusion decision layer concatenates the environmental feature vector and the electrical feature vector, and outputs the false bounce probability after fusion through a fully connected layer.

[0030] Regarding the input layer, for example, it serves as the interface between the model and external data. It receives all circuit breaker-related parameters obtained in step S100 and performs data preprocessing, including standardization (e.g., unifying parameters of different magnitudes to the same numerical range to avoid magnitude differences affecting model training), missing value imputation (e.g., filling in blank data from briefly offline sensors), and time alignment (ensuring that the timestamps of electrical parameters match those of environmental parameters, e.g., correlating current data with temperature and humidity data within the same time period). The processed data will then serve as the input to the feature extraction layer.

[0031] Regarding the feature extraction layer, in some embodiments, environmental parameters (such as temperature, humidity, and dew point temperature) are characterized by both spatial correlation and temporal variation. Here, "space" can be understood as the correlation between multiple variables. The environmental feature extraction module employs a CNN-LSTM structure. CNN is a neural network structure adept at extracting features from a spatial dimension. Because there is a certain "spatial" correlation between multiple environmental parameters (e.g., temperature, humidity, and dew point temperature are collected simultaneously on the same device), CNN is used to discover local patterns among them. CNN is used to extract features in local space, such as the correlation between temperature and humidity, and the gradient of dew point temperature changes.

[0032] LSTM is a special type of recurrent neural network (RNN) that can remember past information and determine which information is useful for the current state and which can be forgotten. LSTM has three gate mechanisms: the forget gate (determines which information needs to be discarded from the memory unit), the input gate (determines how much new information from the current input should be added to the memory), and the output gate (determines how the current memory state affects the output). If humidity continuously rises while temperature drops over a period of time, causing the dew point temperature to approach the current ambient temperature, this trend may trigger internal condensation; LSTM can capture this slowly evolving trend.

[0033] In the CNN-LSTM architecture, the data at each time step is first processed by CNN to extract the local spatial features at that moment; then, these features extracted by CNN are input into LSTM in chronological order, and LSTM models the changes of these features over time; the final output is a high-level semantic feature modeled by spatiotemporal joint modeling.

[0034] In some embodiments, for electrical parameters (such as current, voltage, and frequency), the core characteristic is strong time-series nature; therefore, the electrical feature extraction module adopts an LSTM structure. LSTM can effectively model the trend of electrical signals changing over time, such as current fluctuation patterns, abnormal voltage drops or abrupt changes. The module outputs an electrical feature vector, which contains key dynamic behavior information extracted from the electrical parameters, such as whether short-term overload or current distortion occurs.

[0035] Regarding the role of the fusion decision layer, it combines environmental feature vectors with electrical feature vectors to comprehensively determine whether the trip is non-faulty; it concatenates electrical feature vectors and environmental feature vectors along the feature dimension to form a fused comprehensive feature vector, which is then input into a classifier composed of several layers of fully connected neural networks; the network learns the complex interaction between electrical and environmental factors; the output is a value between 0 and 1, representing the probability that the trip is a false trip: close to 0 indicates that the trip is most likely caused by a real fault; close to 1 indicates that the trip is a false trip caused by environmental interference.

[0036] In some embodiments, the pre-training process of the non-faulty trip identification model includes, for example, the following steps:

[0037] 1. Training data collection and preprocessing

[0038] First, a large amount of historical data needs to be collected from circuit breakers and related sensors, including electrical parameters (such as current, voltage, frequency, etc.) and environmental parameters (such as temperature, humidity, dew point temperature, etc.); trip type labels, including trip types marked by humans or expert systems, are also required, categorized into actual fault trips and non-fault false trips. This data should cover all possible scenarios, including normal operation, actual faults, and known false trip events. The collected data should be cleaned to remove outliers and missing values, ensuring data consistency and accuracy.

[0039] 2. Model Training

[0040] Define a loss function, for example, using the cross-entropy loss function to measure the difference between the model's predicted false jump probability and the true label:

[0041] L=-[y·log(p)+(1-y)·log(1-p)]

[0042] Where y is the true label (1 for false tripping, 0 for normal tripping), and p is the probability output predicted by the model.

[0043] The training process of the model includes initializing the model parameters; performing forward propagation using the training set to generate prediction results; calculating the loss value; calculating the gradient using the backpropagation algorithm; updating the model weights using an optimizer (such as Adam or SGD); and, to prevent overfitting, introducing techniques such as L2 regularization and Dropout, and evaluating the performance on the validation set after each iteration (epoch).

[0044] Step S300: Process the circuit breaker tripping based on the false tripping probability, including triggering a delayed reset of the circuit breaker, forcibly disconnecting, or outputting a processing prompt.

[0045] In some embodiments, the circuit breaker tripping is processed based on the false tripping probability. For example, a mapping relationship between probability intervals and processing strategies is constructed, as shown in the table below:

[0046]

[0047] When the false trip probability is significantly higher than the threshold (e.g., 0.8), it is determined that the trip was caused by non-fault factors (e.g., condensation, transient electromagnetic interference), and "delayed reset" is initiated to restore power supply.

[0048] When the probability is in a fuzzy range, the model cannot definitively determine "false trip / real fault," requiring manual intervention to incorporate more multi-dimensional information (such as track coupling anomalies and historical equipment faults) to assist in decision-making. For example, the alert could be sent to the train's intelligent operation and maintenance terminal: "[Circuit breaker tripping event, false trip probability 0.52, recommended checks: 1. Contact network voltage fluctuations 2. Circuit breaker mechanical condition]," with a visual interface displaying key features, such as:

[0049] Environmental characteristics: Humidity rise rate 0.3% / min (high), dew point temperature difference 2℃ (critical).

[0050] Electrical characteristics: Current fluctuation standard deviation 3A (normal), voltage fluctuation 0.5kV (abnormal).

[0051] Historical data: Number of actions increased by 12% in the past week (worthy of attention).

[0052] When the probability of false tripping is extremely low (e.g., below 0.2), the tripping is determined to be caused by a real fault (e.g., short circuit, insulation breakdown). The circuit breaker is kept in the open state to prevent the fault from escalating. For example, if the circuit breaker trips while the train is stopped at a station, the current waveform shows a continuous 2000A overload (far exceeding the rated value), and the voltage drops sharply by 10kV, with a false tripping probability P = 0.15. The circuit breaker is forcibly disconnected, triggering a train fault alarm to prevent a fire caused by a short circuit in the traction network. Simultaneously, fault data is recorded for ground maintenance analysis.

[0053] Optionally, the overall processing logic after the circuit breaker trips is as follows: Figure 3 As shown,

[0054] 1. Handling of high-probability false jumps (P≥0.8):

[0055] Automatically triggers a delayed reset, waiting for environmental conditions to improve (such as a reduction in the risk of condensation).

[0056] Before resetting, check the stability of electrical parameters to avoid closing the circuit breaker before the fault is cleared.

[0057] 2. Handling low-probability false jumps (P≤0.2)

[0058] Directly force disconnect and lock the tripped state

[0059] Triggering multiple alarm levels (train control center, ground maintenance)

[0060] Record fault waveform data for post-incident analysis.

[0061] 3. Handling of uncertain intervals (0.2) <P<0.8)

[0062] Output processing tips with detailed characteristics (such as humidity rise rate and voltage fluctuation value).

[0063] Human intervention in decision-making, retaining ultimate control.

[0064] System records are updated synchronously after executing manual commands.

[0065] Optionally, based on the LSTM-GARCH sub-model, the stability of electrical parameters within a preset time after tripping is predicted, and a state stability score is output; based on the state stability score S, the weight coefficient k of the false trip probability P1 is dynamically adjusted, and the final false trip probability P = P1 × k is calculated.

[0066] In some embodiments, relying solely on the "current false trip probability P1" after a circuit breaker trips may not be comprehensive enough in certain scenarios. For example, during train operation, electrical parameters change dynamically. In complex conditions, such as the complex electromagnetic environment inside mountain tunnels, even if the current false trip probability P1 indicates a certain possibility of a false trip after the circuit breaker trips, the recovery of electrical parameters may take time. If only the current P1 is considered, inaccurate judgments may be made before the electrical parameters have fully demonstrated a recovery trend. For instance, after a sudden current surge, it may be difficult to determine whether it is a false trip or a potential fault in a very short time. The current may appear stable for 1-2 seconds after the trip, but secondary fluctuations may occur subsequently. The current false trip probability P1 is calculated based only on the parameter extraction features at the moment of tripping and before, without considering the dynamic changes in subsequent parameters.

[0067] Meanwhile, the train operating environment is constantly changing, and the impact of environmental parameters on circuit breaker tripping is also dynamic. For example, in a high-humidity environment, condensation may not form instantaneously. The current false trip probability P1 calculated from environmental parameters such as humidity and dew point temperature difference at the time of tripping does not represent the changes in the environment after tripping. If humidity continues to rise after tripping, the risk of condensation will further increase. Even if P1 at the moment of tripping shows a low false trip probability, subsequent environmental factors may cause the actual false trip probability to increase.

[0068] Furthermore, the current false trip probability P1 is calculated based on electrical and environmental parameters before and at the moment of tripping, through a feature extraction layer and a fusion decision layer. However, these feature extractions are based on data within a finite time window, and cannot capture new features that may appear after tripping in a timely manner. For example, a brief voltage harmonic interference may occur after tripping. This interference does not exist at the moment of tripping, but it will affect the operating state of the circuit breaker. The existing model for calculating P1 cannot take into account this new interference feature that appears after tripping.

[0069] To address the aforementioned potential problems, this embodiment simultaneously predicts whether the electrical parameters will remain stable within a very short timeframe (e.g., 10 seconds) to help correct the probability of false tripping. An LSTM-GARCH sub-model predicts the stability of the electrical parameters within a preset timeframe after tripping and outputs a state stability score.

[0070] Specifically, the LSTM-GARCH sub-model structure includes an LSTM module and a GARCH module.

[0071] For the LSTM module:

[0072] Input: Electrical parameters continuously collected after the trip (e.g., collected every 0.1 seconds, for a total of 100 data points, forming a sequence X). seq =[I1,V1,f1,I2,V2,f2,…,I 100 V 100 ,f 100 ]);

[0073] It is used to learn how parameters change over time, such as whether the current rises sharply and then falls back quickly (false tripping characteristic) or continues to rise (fault characteristic); whether the voltage fluctuates briefly and then stabilizes or oscillates continuously.

[0074] Output: Time series prediction results (Predicting electrical parameters for future, such as multiple time steps).

[0075] For the GARCH module, the GARCH model is a generalized autoregressive conditional heteroskedasticity model:

[0076] Input: The prediction residual (actual value - predicted value) of the LSTM output, or the fluctuation sequence directly based on the original parameters.

[0077] Used to analyze the fluctuation characteristics of the predicted residual (actual value - predicted value) through the GARCH model, and to capture fluctuation clustering phenomena (such as large fluctuations being followed by large fluctuations). For example, is the current fluctuation rate low and stable (recovering to normal after a false trip) or high and persistent (fault not eliminated)? A sudden increase in voltage fluctuation rate may indicate a contact network fault.

[0078] Key steps include:

[0079] Residual calculation: measures the prediction error of LSTM;

[0080] GARCH modeling: The structure is as follows: GARCH(1,1), and the formula is:

[0081] Where w is the long-term average volatility, representing the long-term average volatility level of electrical parameters, reflecting the inherent volatility (such as voltage fluctuations during normal operation of the overhead contact line), and is a constant term in the model, which can be obtained by fitting historical data (such as regression of voltage fluctuation data from the past month).

[0082] The squared prediction residual at time (t-1) represents the degree of deviation between the LSTM model's predicted value and the actual value, reflecting the impact of recent fluctuations (large residuals → large fluctuations); it can be simplified to the formula:

[0083]

[0084] Among them, X t-1 These are the actual electrical parameters. These are LSTM predictions.

[0085] Here, α is the weight of the residual term, which measures the degree of influence of recent volatility on current volatility and reflects the clustering of volatility (large residual → large future volatility). It can be fitted based on historical data, and is usually 0 < α < 1.

[0086] in, Let represent the volatility at time t-1, and let represent the conditional volatility at time t-1. Let represent the continued impact of historical volatility (volatility has memory properties). Let be a recursive term in the model, calculated from the volatility at the previous time step.

[0087] Here, β represents the weight of volatility, which measures the degree of influence of historical volatility on current volatility and reflects the persistence of volatility (large historical volatility → large future volatility). It is fitted with historical data, and usually 0 < β < 1, and α + β < 1 ensures stability.

[0088] Output: Volatility forecasting refers to the degree of fluctuation of electrical parameters (such as current and voltage) at time t, given known historical information, and represents the predicted volatility value at a future time (such as t seconds after a trip).

[0089] Using three parameters—w (long-term average), α (recent residual), and β (historical fluctuation)—the GARCH(1,1) model accurately characterizes the long-term trend, short-term shocks, and persistence of electrical parameter fluctuations.

[0090] Specifically, the state stability score S essentially integrates the "time series recovery trend" (LSTM module) and the "fluctuation risk" (GARCH module) to quantify the health of electrical parameters after a trip. The calculation of the state stability score S can be simplified to the following formula:

[0091] S = α·LSTM score +(1-α)·(1-GARCH volatility )

[0092] Among them, LSTM score This indicates the degree of fit between the electrical parameters (such as current I and voltage V) predicted by LSTM and their actual values. Taking current as an example, let's assume the actual value is I. reated=200A, LSTM predicts the current sequence for the next 5 seconds as follows: The goodness of fit is then: Similarly, for voltage and frequency, the final LSTM... score Take a weighted average of multiple parameters (e.g., current weight 0.5, voltage weight 0.3, frequency weight 0.2); if LSTM score =0.9 indicates that the deviation between the predicted parameters and the actual values ​​is only 10%, showing a good recovery trend; if LSTM score =0.3 indicates that the predicted parameter continues to deviate from the actual value, indicating a poor recovery trend.

[0093] Among them, GARCH volatility This represents volatility prediction, specifically the volatility of electrical parameters (such as voltage standard deviation and current fluctuation amplitude) predicted by the GARCH model, measuring the stability of the parameter regression process. The future volatility calculated using GARCH(1,1) is normalized (e.g., divided by 5% of the nominal value as the maximum volatility threshold) to obtain the GARCH value. volatility If GARCH volatility =0.1 indicates low volatility, representing a stable recovery process; if GARCH volatility =0.8 indicates high volatility, suggesting an unstable recovery process.

[0094] Here, α represents the weight of the weight coefficient that balances the "time series recovery trend" and "volatility risk", reflecting the model's emphasis on the two types of features.

[0095] Specifically, the weight coefficient k is dynamically adjusted based on the calculated state stability score S. In this embodiment, the logic for dynamic adjustment is as follows:

[0096] If the electrical parameters recover well and fluctuate little after tripping (S is high) → the possibility of false tripping is high, so increase k (to make the final probability P = P1 × x higher);

[0097] If the electrical parameters recover poorly and fluctuate greatly after tripping (low S), the probability of a real fault is high, so reduce k (to make the final probability P even lower).

[0098] For example, the rules for segmented adjustment are shown in the table below:

[0099]

[0100] When S≥0.8, the LSTM prediction parameters quickly return to the rated range, and the GARCH prediction volatility continues to decrease. This type of scenario is highly consistent with the characteristics of false bounce (instantaneous fluctuations caused by non-fault factors, followed by rapid recovery). The weight of the false bounce probability needs to be increased, and a delayed reset is preferred.

[0101] When 0.2 < S < 0.8, the LSTM prediction parameter shows a recovery trend but the fluctuation has not completely converged, and it is still higher than the rated value, and the GARCH predicted volatility maintains a medium level; such scenarios belong to edge cases (which may be false tripping or may be actual faults), it is necessary to maintain the original probability weight of false tripping and trigger a prompt for manual intervention.

[0102] When S ≤ 0.2, the LSTM prediction parameter continuously deviates from the rated value, and the GARCH predicted volatility increases. Such scenarios are highly consistent with the characteristics of real faults (e.g., continuous parameter abnormality caused by short circuit), it is necessary to reduce the weight of false tripping probability and prefer forced tripping.

[0103] By dynamically adjusting k through S, the dynamic correction of false tripping probability from static calculation is realized, which makes the tripping treatment strategy (delayed reset / forced disconnection) more consistent with the actual recovery state of the electrical system; it not only reduces operation interruption caused by false tripping, but also prevents the risk of fault expansion.

[0104] Optionally, said performing stability prediction on electrical parameters within a preset time after tripping further comprises: arranging a plurality of time windows with different lengths for parallel prediction in parallel, dynamically determining the weights of features of different windows through an attention mechanism, and finally fusing them into said state stability score.

[0105] In the previous embodiment, there are some problems in using a fixed time window (e.g., 10 seconds after tripping) to predict stability. For example, the recovery periods of different faults vary greatly (e.g., condensation-induced false tripping recovers quickly, while short-circuit fault recovers slowly), and the fixed window may miss key features; it is impossible to dynamically distinguish the importance of short-term fluctuation (e.g., sudden current rise within 0.1 seconds) and long-term trend (e.g., voltage recovery within 10 seconds).

[0106] To solve the above problems, in this embodiment, a plurality of time windows with different lengths are arranged in parallel for parallel prediction, the weights of features of different windows are dynamically determined through a gating unit, and finally fused into said state stability score. Specifically, time windows are classified according to time length, which is exemplarily shown as follows:

[0107]

[0108] Each window runs the LSTM-GARCH model independently to obtain the feature vector h reflecting the time-series recovery trend (output by LSTM) and fluctuation characteristics (output by GARCH) s ,h m ,h l (respectively represent the feature vectors of short-term window, medium-term window and long-term window).

[0109] Map the feature vectors of each window to a unified semantic space, h' s =W s h s +bs ,h' m =W m h m +b m ,h' l =W l h l +b l ;in and d represents the feature dimension.

[0110] The importance weights of each window are calculated by scaling the dot product attention, which is used to dynamically evaluate the importance of features in different time windows and assign corresponding weights; this can be simplified to the following formula:

[0111]

[0112] Where Q represents the query vector used to match each window feature to determine which features are more relevant to stability evaluation, and its dimension is d (consistent with the window feature dimension), i.e. K represents the identifier of each window feature, used to calculate the similarity with the query vector Q, measuring feature relevance. The matrix composed of the feature vectors of each window has a dimension of 3×d (3 windows, each window's feature dimension d), i.e. K stores the feature identifiers of short-term, medium-term, and long-term windows for Q query matching. V represents the actual content of each window feature, corresponding one-to-one with the key matrix K, and is used for weighted output. V stores the actual feature values ​​of the short-term, medium-term, and long-term windows for final weighted fusion. As a scaling factor, it addresses the issue of excessively high dot product similarity leading to the vanishing softmax gradient when the feature dimension d increases, resulting in a smoother attention distribution. This ensures that the attention weights can still effectively distinguish the importance of different windows even with high-dimensional features.

[0113] Softmax is a normalization function that transforms the similarity between the query and the key into a probability distribution (with weights summing to 1), giving higher weights to windows with higher importance. The row normalization of the similarity matrix is ​​performed using the following formula:

[0114]

[0115] For example, if a short-term window feature is a better match (higher similarity) to Q, then softmax will give it a higher weight.

[0116] The weight values ​​for each window can be obtained through calculation, namely α. s α m α l ;where α s α m α l ∈[0,1], and αs +α m +α l =1.

[0117] Attention weights are applied to the original feature vector to obtain the fused features, h. fused =α s h s +α m h m +α l h l This vector integrates feature information from different time scales, and its weights are dynamically adjusted by the model based on the current input.

[0118] The fused features are mapped to the final state stability score S through a fully connected layer, where S = σ(W out h fused +b out ), where W out As the weight, b out The bias term is obtained through model self-learning, and σ is the sigmoid activation function, which constrains the output to the interval [0,1].

[0119] Optionally, a penalty term is added to the conditional variance equation of the LSTM-GARCH model to suppress the impact of high-frequency interference on volatility calculation.

[0120] In some embodiments, high-frequency interference (such as electromagnetic radiation or sampling noise) can cause transient spikes or glitches in the parameter sequence (such as small, meaningless oscillations in current over a short period). These interferences can affect the conditional variance equation of the GARCH model. Overfitting to noise, for example, residual terms Containing a large amount of interfering information, leading to volatility prediction It has been falsely amplified.

[0121] To address the aforementioned issues, specifically to suppress high-frequency interference, a regularization penalty term (such as L1 / L2 regularization) is added to the GARCH conditional variance equation. The modified equation takes the following form:

[0122]

[0123] Where θ = {w, α, β} are model parameters, λ ≥ 0 is the penalty strength, and Penalty(θ) is the regularization function (such as the L2 penalty term).

[0124] High-frequency disturbances are characterized by being short-term, highly volatile, and lacking persistence. This causes the GARCH parameter α (residual term weight) to abnormally increase in order to fit the noise. The penalty term, by constraining the range of α (e.g., L2 penalty to bring α closer to 0), weakens the excessive influence of the residual term on volatility, ensuring the model focuses on the persistent volatility of real faults (e.g., persistent current anomalies caused by short circuits). The volatility of real faults has temporal persistence (e.g., current fluctuations in short-circuit faults last for multiple time steps), while the volatility of high-frequency disturbances is instantaneous and isolated. The penalty term reinforces the impact of volatility persistence by adjusting β (historical volatility weight): if β increases due to the penalty term, the model will focus more on long-term volatility trends rather than short-term noise interference.

[0125] By constraining model parameters, the penalty term helps GARCH focus more on the persistent fluctuation characteristics of real faults (such as the persistent current anomaly in short-circuit faults) rather than transient noise, making the state stability score S more accurately reflect the actual recovery state of the circuit breaker. This suppresses the overfitting of high-frequency disturbances to the residual term and historical volatility; enhances the persistent characteristics of real fault fluctuations; and ultimately improves the accuracy of the state stability score S.

[0126] Optionally, the model also includes acquiring the circuit breaker's relevant state parameters and adding a state parameter input branch to the LSTM-GARCH model. In the LSTM-GARCH model of the aforementioned embodiments, the main inputs are the electrical parameters (current, voltage, frequency) after tripping. However, in real-world scenarios, the circuit breaker's historical behavior (opening and closing records) and system correlations (signals from adjacent devices) are equally crucial for judging false trips: opening and closing records can reflect the long-term health status of the equipment (e.g., frequent actions may indicate mechanical faults); signals from adjacent devices can reflect the propagation of system-level faults (e.g., pantograph malfunctions may cause the circuit breaker to trip, rather than it tripping itself).

[0127] The status parameters related to the circuit breaker mainly include the circuit breaker status change record, such as the opening and closing timestamp, trip response time (e.g., the time taken from the fault signal to the contact separation), to determine whether the mechanical characteristics of the circuit breaker are normal, and the action waveform data record of the real-time waveforms of coil current and mechanical displacement during the opening and closing process, such as the coil current should fluctuate between 50-80A when closing, the cumulative number of opening and closing actions, etc.; as well as the trip signals of adjacent equipment, the opening and closing status of other circuit breakers (such as the traction converter input side circuit breaker), pantograph, surge arrester and other equipment in the same power supply arm.

[0128] In this embodiment, by adding a new status parameter input branch, two types of key information are supplemented: opening and closing records, which determine whether the tripping is an accidental malfunction or a continuation of equipment failure by using the equipment's historical operation mode; and adjacent equipment signals, which analyze the system's correlation effects and determine whether the tripping is caused by its own malfunction or by an external fault (such as the contact network or pantograph).

[0129] Specifically, regarding the characteristics of the opening and closing records:

[0130] By analyzing historical opening and closing events using a sliding window, the following features are extracted (taking a train circuit breaker as an example): Recent operation frequency: the number of opening and closing events in the past hour (normal ≤ 10 times, abnormal > 20 times); Longest interval time: the interval between the two most recent opening and closing events (normal ≥ 30 minutes, abnormal ≤ 5 minutes); Operation time distribution: the proportion of operations during peak hours (such as when trains start and stop).

[0131] Specifically, regarding the signal characteristics of adjacent devices:

[0132] Collect the status sequence of adjacent equipment within 10 seconds before and after the trip, and extract the following features (taking pantograph and adjacent circuit breaker as examples): Pantograph voltage fluctuation: voltage curve of pantograph before and after tripping (such as whether there is a sudden drop or harmonics); Status of adjacent circuit breakers: whether other circuit breakers operate synchronously (such as "when circuit breaker A trips, circuit breaker B also trips" may indicate a power grid fault); Equipment coordination mode: the timing correlation between the operation of pantograph and circuit breaker (such as the circuit breaker tripping 0.1 seconds after pantograph goes offline).

[0133] The newly added status parameters (opening and closing records, adjacent equipment signals) are input into the LSTM-GARCH model in parallel with the existing electrical parameters (current, voltage), and then fused in the following way:

[0134] Opening and closing record characteristics → input the historical behavior branch of the LSTM to learn the long-term stability of the device;

[0135] Adjacent device signals are input into the system correlation branch of the LSTM to learn external fault propagation.

[0136] Electrical parameters → Input the existing LSTM-GARCH branch to learn instantaneous fluctuations and recovery trends.

[0137] By supplementing the state parameters, the state stability score S of the LSTM-GARCH model is more comprehensive, improving the reliability of S and making the tripping handling strategy (delayed reset / forced disconnection) more realistic.

[0138] Figure 4 An intelligent control system 400 for a train circuit breaker is shown. An embodiment of this system is described. Figure 1 Corresponding to the method embodiments shown, the system can be specifically applied to various electronic devices.

[0139] like Figure 4 As shown in the embodiment of this application, the intelligent control system 400 for train circuit breakers includes:

[0140] The first acquisition module 401 is used to acquire circuit breaker-related parameters in response to circuit breaker tripping. The circuit breaker-related parameters include electrical parameters, environmental parameters, and status parameters.

[0141] The first prediction module 402 is used to input the relevant parameters of the circuit breaker into the non-fault trip identification model, and predict the false trip probability of the circuit breaker based on the identification model. The non-fault trip identification model includes an input layer, a feature extraction layer and a fusion decision layer. The input layer is used to receive the relevant parameters of the circuit breaker. The feature extraction layer outputs a condensation risk feature vector and an electrical anomaly feature vector. The fusion decision layer outputs the false trip probability.

[0142] The processing module 403 is used to process the circuit breaker tripping based on the false tripping probability, including triggering a delayed reset of the circuit breaker, forcibly disconnecting, or outputting a processing prompt.

[0143] Based on the same inventive concept, this application also provides an electronic device. The method corresponding to the electronic device can be the method in the foregoing embodiments, and its problem-solving principle is similar to that method. The electronic device provided in this application includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the methods and / or technical solutions of the foregoing embodiments of this application.

[0144] The electronic device can be a user device, or a device formed by integrating user devices and network devices through a network, or it can be an application running on the aforementioned devices. The user device includes, but is not limited to, various terminal devices such as computers, mobile phones, tablets, smartwatches, and wristbands. The network device includes, but is not limited to, network hosts, single network servers, multiple network server sets, or cloud computing-based computer sets, and can be used to implement some processing functions when setting an alarm clock. Here, the cloud consists of a large number of hosts or network servers based on cloud computing. Cloud computing is a type of distributed computing, consisting of a virtual computer composed of a group of loosely coupled computer sets.

[0145] Figure 5The diagram illustrates the structure of an apparatus suitable for implementing the methods and / or technical solutions in the embodiments of this application. The apparatus 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes based on a program stored in a read-only memory (ROM) 502 or a program loaded from a storage portion 508 into a random access memory (RAM) 503. The RAM 503 also stores various programs and data required for system operation. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0146] The following components are connected to I / O interface 505: input section 506 including keyboard, mouse, touch screen, microphone, infrared sensor, etc.; output section 507 including cathode ray tube (CRT), liquid crystal display (LCD), LED display, OLED display, etc., and speakers, etc.; storage section 508 including one or more computer-readable media such as hard disk, optical disk, magnetic disk, semiconductor memory, etc.; and communication section 509 including network interface card such as LAN (local area network) card, modem, etc. Communication section 509 performs communication processing via a network such as the Internet.

[0147] In particular, the methods and / or embodiments in this application can be implemented as computer software programs. For example, the embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowchart. When the computer program is executed by the central processing unit (CPU) 501, it performs the functions defined in the methods of this application.

[0148] Another embodiment of this application provides a computer-readable storage medium having computer program instructions stored thereon, which can be executed by a processor to implement the methods and / or technical solutions of any one or more embodiments of this application described above.

[0149] Specifically, this embodiment may employ any combination of one or more computer-readable media. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

[0151] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a device claim may also be implemented by a single unit or device through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A method for intelligent control of a train circuit breaker, characterized by, include: In response to a circuit breaker tripping, relevant parameters of the circuit breaker are acquired, including electrical parameters and environmental parameters; The circuit breaker-related parameters are input into a non-fault tripping identification model. Based on the identification model, the false tripping probability of the circuit breaker is predicted. The non-fault tripping identification model includes an input layer, a feature extraction layer, and a fusion decision layer. The input layer receives the circuit breaker-related parameters. The feature extraction layer outputs environmental feature vectors and electrical feature vectors. The fusion decision layer outputs the false tripping probability. The feature extraction layer includes an environmental feature extraction module and an electrical feature extraction module. The environmental feature extraction module uses a CNN-LSTM structure to output environmental features containing spatiotemporal information. The electrical feature extraction module uses an LSTM structure to learn the temporal dependencies of the electrical parameters and output electrical features. The fusion decision layer concatenates the environmental features and electrical features, fuses them through a fully connected layer, and outputs the false tripping probability. Based on the LSTM-GARCH sub-model, the stability of the electrical parameters within a preset time after tripping is predicted, and a state stability score is output. The weight coefficients of the false tripping probability are dynamically adjusted based on the state stability score to calculate the final false tripping probability. The circuit breaker tripping is processed based on the false tripping probability, including triggering a delayed reset of the circuit breaker, forcibly disconnecting, or outputting a processing prompt.

2. The intelligent control method of a train circuit breaker according to claim 1, wherein, Also includes: The method of predicting the stability of electrical parameters within a preset time after tripping also includes setting multiple time windows of different lengths in parallel for prediction, dynamically determining the weights of different window features through an attention mechanism, and finally merging them into the state stability score.

3. The intelligent control method of a train circuit breaker according to claim 2, wherein, Also includes: A penalty term is added to the conditional variance equation of the LSTM-GARCH sub-model to suppress the impact of high-frequency interference on volatility calculation.

4. An intelligent control system for train circuit breakers, characterized in that, include: The first acquisition module is used to acquire circuit breaker-related parameters in response to circuit breaker tripping. The circuit breaker-related parameters include electrical parameters, environmental parameters, and status parameters. The first prediction module inputs the circuit breaker-related parameters into a non-fault tripping identification model, and predicts the false tripping probability of the circuit breaker based on the identification model. The non-fault tripping identification model includes an input layer, a feature extraction layer, and a fusion decision layer. The input layer receives the circuit breaker-related parameters, the feature extraction layer outputs environmental feature vectors and electrical feature vectors, and the fusion decision layer outputs the false tripping probability. The feature extraction layer includes an environmental feature extraction module and an electrical feature extraction module. The environmental feature extraction module uses a CNN-LSTM structure to output environmental features containing spatiotemporal information. The electrical feature extraction module uses an LSTM structure to learn the temporal dependencies of the electrical parameters and output electrical features. The fusion decision layer concatenates the environmental features and electrical features, fuses them through a fully connected layer, and outputs the false tripping probability. Based on the LSTM-GARCH sub-model, the stability of the electrical parameters within a preset time after tripping is predicted, and a state stability score is output. The weight coefficients of the false tripping probability are dynamically adjusted based on the state stability score to calculate the final false tripping probability. The processing module is used to process the circuit breaker tripping based on the false tripping probability, including triggering a delayed reset of the circuit breaker, forcibly disconnecting, or outputting a processing prompt.

5. An electronic device, the electronic device comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, characterized in that: the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1-3.

6. A computer-readable medium having computer program instructions stored thereon, characterized in that: The computer program instructions can be executed by a processor to implement the method as described in any one of claims 1-3.

7. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, it implements the method as described in any one of claims 1-3.

Images